It has nothing to do with fine-tuning - like most of these startups using the openai api. It's basically all clever prompt engineering techniques that emerged over the last years finally combined into ChatGPT.